In recent years, a user can enjoy a desired content chosen from among an enormous number of contents using a communication network such as the Internet. Since there are an enormous number of available contents, various kinds of recommendation techniques have been suggested. For example, a technique for calculating a degree of similarity between a preference vector representing a feature of the content preferred by an user, and a feature vector of each of contents while recommending the user the content with the high degree of similarity, so called content-based filtering is one of the examples (see Japanese Patent Application Laid-Open No. 2001-160955).